We followed recent results indicating differences in the density and size of human and mouse supragranular pyramidal cells (PCs) (Berg et al., 2021) in a human-rat setting. As expected, measurements on 3D reconstructions based on randomly selected, electrophysiologically recorded and biocytin filled human (n = 30) and rat (n = 30) L2/3 cortical pyramidal cells (Fig. S1A) show significant differences in the horizontal (463.17 ± 119.48 vs. 324.79 ± 80.58 µm, t test: P = 1.687 × 10^-6) and vertical extensions (542.58 ± 146.89 vs. 409.99 ± 102.69 µm, t test: P = 0.00013), and in the total dendritic (9054.94 ± 3699.71 vs. 5162.68 ± 1237.71 µm, t test: P = 7.203 × 10^-7) and apical dendritic length (4349.76 ± 1638.39 vs. 2592.15 ± 818.26 µm, t test: P = 1.638 × 10^-6, Fig. S1B,C).

To examine the temporal aspects of information propagation in excitatory microcircuits, we performed simultaneous whole cell patch clamp recordings in synaptically connected L2/3 PCs from acute neocortical slices from rat and human tissues (Fig. 1). Excitatory postsynaptic potentials (EPSPs) were measured in response to single action potentials (AP) in presynaptic cells (Fig. 1B). Synaptic latency was calculated as the time difference between the peak of the presynaptic AP and the onset point of the postsynaptic EPSP (see Fig 1B and Methods). We did not find significant differences in synaptic latencies between human and rat PC-to-PC connections (rat: 1.126 ± 0.378 ms, rat: n = 19, human: 1.111 ± 0.306 ms, n = 17, Mann-Whitney test: P = 0.949). Both pre- and postsynaptic PCs were filled with biocytin during recordings allowing for post hoc identification of close appositions between presynaptic axons and postsynaptic dendrites (Frick et al., 2008) (Fig. 1A). We measured the shortest axonal path lengths linking the presynaptic soma to close appositions on the postsynaptic dendrite (rat: 168.267 ± 49.59 µm, human: 272.22 ± 73.14 µm) and the shortest dendritic path lengths from close appositions found exclusively on dendritic spine heads to the postsynaptic soma (rat: 84.9 ± 18.301 µm, human: 129.48 ± 40.005 µm) in a subset of recordings (rat: n = 6, human: n = 5). Consequently, we found that the minimal intersomatic distance (the sum of the shortest axonal and dendritic paths) in each synaptically connected PC-to-PC pair was significantly smaller in rats compared to humans (rat: 259.7 ± 58.8 µm, human: 402.12 ± 74.757 µm, Mann-Whitney test: P = 0.009, Fig. 1D). We did not find significant difference in these paired recordings in synaptic latency (rat: 1.09 ± 0.375 ms, n = 6 from n = 6 rats; human: 1.102 ± 0.408 ms, n = 5 from n = 5 patients; Mann-Whitney test: P = 0.931, Fig. 1C, darker dots). Given that similar synaptic latencies accompany different lengths for signal propagation in the two species, membrane potentials (APs and/or EPSPs) are likely to propagate faster in human PC-to-PC connections.

Compensation of longer axonal and dendritic paths must be explained by higher velocity of signal propagation along axons and/or dendrites. We therefore asked whether interspecies differences can be found in axonal and/or dendritic signal propagation in L2/3 PCs.

First, we investigated whether we could find dissimilarities between the two species in the speed of signal propagation along axons of PCs. We whole cell recorded the soma and a distal axon simultaneously, positioning the axonal recording electrode on one of the blebs formed at the cut ends of axons during slice preparation. Somatic current injections were used to trigger APs and the time between somatic and the axonal AP was measured (Fig. 2A). We captured two-photon images during electrophysiological recording and measured the length of the axonal path from the somatic to the axonal electrode on image z-stacks. The dataset was restricted to recordings that matched the distances from the soma to axo-dendritic close appositions determined above along the axon of synaptically coupled PC-to-PC connections (rat: n = 8, 268.203 ± 76.149 µm vs. human: n = 9, 281.507 ± 125.681 µm, two sample t test: P = 0.799, Fig. 2F). The latency between the soma and the axon bleb of the propagating AP peaks was not significantly different between the species (rat: n = 8, 0.333 ± 0.211 ms vs. human: n = 9, 0.327 ± 0.123 ms, two sample t test: P = 0.945). The axonal speed of AP propagation was calculated for each cell from the time required from soma to recording site. We found no significant difference in the propagation speed of APs in the axons of rats and humans (rat: n = 8, 0.848 ± 0.291 m/s vs. human: n = 9, 0.851 ± 0.387 m/s, two sample t-test: P = 0.282, Fig. 2F). Our axonal recordings suggest that there is no significant difference between the two species over the range of distances we investigated, so the lower latencies in the paired recordings may be due to dendritic differences.

We next sought to test rat and human dendritic signal propagation velocity using simultaneous whole cell patch clamp recordings with electrodes placed on the somata and dendritic shafts of PCs. Distances of somatic and dendritic recording locations (rat: 143.078 ± 72.422 µm, n = 46; vs. human: 153.446 ± 57.698 µm, n = 62, Mann-Whitney test: P = 0.175, Fig.2B) were chosen to be similar in the two species and in range of soma-to-dendrite distances of axo-dendritic close appositions determined above for synaptically coupled PC-to-PC connections. In the first set of experiments, we injected suprathreshold current through the somatic electrode and measured the time difference between the evoked AP peak at the soma and the respective backpropagating AP peak in the dendritic electrode (Fig. 2E and F). We found significant difference in the signal propagation time between rat and human PCs (rat: 0.672 ± 0.334 ms, n = 46; vs. human: 0.495 ± 0.229 ms, n = 62, Mann-Whitney test: P = 0.005, Fig. 2F). The AP propagation speed was calculated for each cell from the time difference between the somatic and dendritic APs divided by the distance between the two points. We found that the propagation speed was, on average, ∼1.47-fold faster in human (rat: 0.233 ± 0.095 m/s vs. human: 0.344 ± 0.139 m/s, Mann-Whitney test: P = 6.369 × 10^{- 6}, Fig. 2F). In a second set of experiments, using the same dual recording configuration, we tested orthodromic or forward propagating signal propagation velocity by injecting short-duration current ramps to simulate EPSP (sEPSP) signals in the dendrites and recorded the resultant subthreshold voltage response in the soma (Fig. 2C). These experiments were performed in the same PCs where backpropagating AP velocities were also measured (rat: n = 24, human: n = 24). We found that sEPSP propagation speed was, on average, ∼1.26-fold faster in human (rat: 0.074 ± 0.018 m/s vs. human: 0.093 ± 0.025 m/s, two sample t test: P = 0.004; Fig. 2D). In addition, we found correlation between forward propagating sEPSP speed and back propagating AP speed (Pearson correlation coefficient, r = 0.396, P = 0.0053, Fig. 2D).

Hyperpolarization-activated cyclic nucleotide-modulated (HCN) channel densities were shown to be higher in human compared to rat layer 2/3 PCs and were shown to be instrumental in more depolarized resting membrane potentials and in larger sag potentials in response to hyperpolarization in the human (Kalmbach et al., 2018). In addition, modeling predicted that signal delay in dendrites reduces with increased h-conductance (Kalmbach et al., 2018). In line with previous studies, human PCs in our dataset had more depolarized resting membrane potential (rat: -70.49 ± 5.78 mV, human: -64.30 ± 7.28 mV, Mann-Whitney U test: P = 7.37× 10^-6, Fig. S2A) but the average somatic input resistance were not significantly different in the two species (rat: 59.56 ± 21.86 MΩ, n = 46, human: 71.375 ± 65.485 MΩ, n = 62, Mann-Whitney test: P = 0.347, Fig. S2A).

Based on the correlation found between forward-propagating sEPSP speed and back-propagating AP speed, we performed pharmacological experiments on bAPs (since it is technically less challenging to evoke) to uncover potential contributors to increased dendritic speeds in humans. To test the contribution of h-channels to the elevated signal propagation speed in human dendrites, we performed pharmacological experiments with 20 µM ZD7288, a specific blocker of h-channels. Significant hyperpolarization of the resting membrane potential was observed in the human cells but not in the rat neurons (Fig. S2B) and significantly increased input resistance accompanied drug application in both human and rat neurons (Fig. S2C). Drug administration did not significantly decrease bAP propagation speed in rat PCs (control: 0.163 ± 0.054 m/s, ZD7288: 0.149 ± 0.057 m/s, n = 9, two-way ANOVA with repeated measures and Bonferroni post-hoc correction: P = 1, Fig. 3A) and in human PCs (control: 0.322 ± 0.073 m/s, ZD7288: 0.268 ± 0.066 m/s, n = 8, two-way ANOVA with repeated measures and Bonferroni post-hoc correction: P = 0.932, Fig. 3B). The changes in bAP propagation speed were higher in the human cells (rat: -0.014 ± 0.019 m/s, human: -0.054 ± 0.052 m/s, two-sample t test: P = 0.048, Fig. 3C) in response to h-channel blockage. It can therefore be argued that HCN channels may contribute to the higher conduction velocities in human dendrites, but do not by themselves explain the differences between the two species.

Back-propagation of APs is an active process supported by voltage gated ion channels that can initiate regenerative events in the dendrites (Stuart & Sakmann, 1994). To further investigate the influence of voltage gated ion channels we pharmacologically blocked voltage gated Na⁺ channels with tetrodotoxin (TTX, 1µM), voltage gated Ca²⁺ channels with cadmium chloride (CdCl₂, 200 µM), and NMDA receptors with (2R)-amino-5-phosphonovaleric acid (AP5, 20 µM) simultaneously. Since the blockage of voltage gated Na⁺ channels prevent the initiation of APs, we kept the soma of the recorded cells in voltage clamp mode and used a prerecorded template as voltage command through a somatically placed electrode (the so called “simulated spike”) and measured the back propagation of the response to the somatic voltage command at a dendritic recording site in current clamp mode. As expected, the amplitude of the bAPs at the dendritic recording site dropped significantly in human and rat cells respectively (Fig. S2D). The speed of back propagation of membrane potential signals in dendrites with blocked regenerative events by the pharmacological cocktail was not significantly reduced in rat samples compared to the drug-free control (rat control: 0.199 ± 0.053 m/s, rat TTX/CdCl₂/AP5: 0.076 ± 0.03 m/s, two-way ANOVA with repeated measures and Bonferroni post-hoc correction: P = 0.0587, Fig. 3D), but was significantly lower in human samples (human control: 0.395 ± 0.14 m/s, human TTX/CdCl₂/AP5: 0.184 ± 0.061 m/s, two-way ANOVA with repeated measures and Bonferroni post-hoc correction: P = 0.004, Fig. 3E). The human dendrites with blocked action potential generation maintained a higher bAP propagation speed (rat: 0.076 ± 0.03 m/s n = 8, human: 0.184 ± 0.061 m/s n = 8, two-way ANOVA with repeated measures and Bonferroni post-hoc correction: P = 0.0956 Fig. 3F). In summary, in the search for factors contributing to higher signal propagation speeds in human pyramidal dendrites compared to rat pyramids, voltage-gated Na⁺, Ca²⁺ and NMDA channels appear to play a less role in differentiating the two species, complemented by a minor contribution from HCN channels, which are differentially dense in human vs. rat.

The specific membrane capacitance (C_m) can influence the time constant of the biological membrane, and it is a key determinant of the propagation of electrical signals. Recent experiments indicated that the C_m of human L2/3 PCs might be significantly lower compared to rodents (Eyal et al., 2016) and modeling studies suggested that the decrease in C_m could lead to increased conduction speed and fewer synapses being able to evoke suprathreshold events in human PCs (Eyal et al., 2016). However, a separate line of experiments could not detect differences in the C_m of L5 PCs between humans and rodents (Beaulieu-Laroche et al., 2018), or L2/3 PCs (Gooch et al., 2022) thus, to test whether C_m is a component in producing elevated signal propagation velocity in human dendrites, we directly measured the C_m values of human and rat PCs by pulling nucleated patches (Eyal et al., 2016) (Fig. 4A,B). We found no significant difference in the C_m between the human and rat L2/3 PCs (rat: 1.092 ± 0.14 µF/cm² n = 20; human: 0.987 ± 0.196 µF/cm² n = 19, two-sample t test: P = 0.0615, Fig. 4C). The specific membrane capacitance is determined by the dielectric constant of the membrane, and it is inversely proportional with the membrane thickness. We measured the membrane thickness of dendritic structures with transmission electron microscopy both in human and rat samples (Fig. 4D,E) and detected no significant differences between the two species (human: 4.271 ± 0.873 nm, n = 213 from n = 3 patient; rat: 4.122 ± 0.779 nm n = 151 from n = 3 rat, Mann-Whitney test: P = 0.212, Fig. 4E). Based on these experiments it seems that not the specific membrane capacitance is the key determinant of the higher signal propagation speed in human cells.

The relationship between conduction velocity and axon diameter is well known for small myelinated and unmyelinated axons (Waxman & Bennett, 1972). Anatomical features of neuronal dendrites also have a major influence on signal propagation properties (Deitcher et al., 2017; Rall & Rinzel, 1973; Rinzel & Rall, 1974; Vetter et al., 2001), thus, in addition to the soma-dendritic path measurements shown above, we also measured the thickness of dendrites at every 0.5 µm along the path linking the somatic and dendritic electrodes on two-photon image stacks captured during electrophysiological measurements (Fig. 5A-C). We found that the mean diameter of dendrites was thicker in human (2.272 ± 0.584 µm, n = 62) compared to the rat (2.032 ± 0.413 µm, n = 46, two sample t test: P = 0.019, Fig. 5D). Moreover, in samples where we acquired both dendrite thickness and bAP signal propagation velocity, we found that the mean dendritic diameter between the recording sites was correlated with the speed of backpropagating APs (Fig. 5E).

Detailed compartmental models were utilized to disassemble the effect of various morphological and cable parameters on the latency and velocity of synaptic potential in human and rat L2/3 dendrites. Based on the 3D morphological reconstructions of five human and four rat PCs, we first asked, how dendritic morphological differences per se affect signal propagation, assuming that the cable parameters are identical in all cells (C_m = 1 µF/cm², R_m = 15,000 Ωcm², R_a = 150 Ωcm, Fig. 6). Figure 6A,B shows EPSPs latency and velocity as a function of distance from its dendritic initiation site to the soma. Latency was calculated as the time difference between the peak-times of the local dendritic EPSP and of the resulting somatic EPSP. The dendritic-to-soma latency ranged between 0.1 - 13 ms in rats (red circles) and 0.01 – 25 ms in human (blue circles). The larger maximal latency in human is expected due to the ∼2-folds longer apical dendrite in humans L2/3 neurons (Figure 6A). The respective EPSPs velocity was calculated by dividing the path distance to the soma from the dendritic origin of the EPSP by its latency (Figure 6B). EPSP velocity ranged between 0.02 - 0.09 m/s in rat and 0.01 - 0.48 m/s in human (Figure 6B). The exceptionally large differences in the maximal velocity between human and rat is taken in the Discussion.

We next compared signal propagation in rat and human dendrites, focusing on identical range of dendrite-to-soma distances in which the experiments were performed (27 µm – 289 µm, insets in Figure 6A,B). Towards this end, we computed the mean EPSP latency and velocity as a function of distance from the soma, averaged across different branches at a given distance from the soma (Figures 6A, 6B, lower right and upper right insets). For this experimental range of recordings, EPSP velocity ranged between 0.04 - 0.08 m/s in humans versus 0.03 - 0.05 m/s in rats. EPSP latency ranged between 0.1 - 4.7 ms in humans versus 0.4 - 6.5 ms in rats. These findings demonstrate significantly faster EPSP propagation in humans compared to rats (average latency in humans: 2.45 ± 0.2 ms, n = 5; in rats: 3.3 ± 0.3 ms, n = 4; Mann-Whitney U test: P = 0.03. Average velocity in humans: 0.08 ± 0.005 m/s, n = 5; in rats: 0.05 ± 0.006 m/s, n = 4; Mann-Whitney U test: P = 0.02. See Fig.S13 and Suppl. Table 2).

A possible reason for the smaller latency and larger velocity of EPSPs in human apical dendrites is that they have larger diameter (Figs. 5D and Fig. 6C see also refs. Agmon-Snir & Segev, 1993; Jack et al., 1975). Theory shows that, for an infinitely long passive cylindrical cable, the velocity of passive signals is not constant. It is fast near their site of origin, converging to a value of 2λ/τ away from their initiation site (Agmon-Snir & Segev, 1993; Jack et al., 1975) (λ is the cables’ space constant and τ is its membrane time constant). This means that the latency and velocity of passive signals, when normalized in units of λ/τ, are identical for cylindrical cables with different diameters (see Fig. S4). This is due to the fact that differences in cable diameter are taken into account when normalizing the physical distance, x, by λ (which is ∝ √𝑑, where d is the cable diameter). Hence, if the larger diameter in human dendrites is a key contributor to the enhanced signal velocity in these cells, we expect that the EPSP latency and velocity will converge on similar curves for all cells (rat and human alike) after normalizing the distance in units of λ, and time in units of τ (see Fig. S4). However, albeit such normalization, the velocity is still larger and the latency is shorter in human (compare insets in Fig. 6D,E to Fig. 6A,B, respectively. In this case, the average latency in humans: 0.16 ± 0.01 τ, n = 5; in rats: 0.22 ± 0.02 τ, n = 4; Mann-Whitney U test: P = 0.02. The average velocity in humans: 2.55 ± 0.11 λ/τ, n=5; in rats: 1.73 ± 0.29 λ/τ, n = 4; Mann-Whitney U test: P = 0.02).

To summarize: Scaling dendritic distance in units of λ and time in units of τ did not eliminate the statistically significant differences in EPSP latency and velocity between humans and rats. This raises the question: what factor enhances EPSP propagation speed in human dendrites?

One possibility is that differences in boundary conditions for EPSPs travelling from the dendrites towards the soma might explain the enhanced signal propagation in human. Boundary conditions are known to affect the steepness of voltage attenuation along the dendrites (Rall & Rinzel, 1973; Rinzel & Rall, 1974). But do differences in the boundary condition (“the impedance load”) at the soma affect the speed of EPSPs propagating when travelling from the apical tree towards the soma? Notably, the basal tree in human L2/3 PCs is substantially larger than that of rat (Fig. 6F and Fig. 8A). Quantifying the total membrane area confirms that the basal tree in human is significantly larger than in rats (in humans: 23,621 ± 7,735 µm², n = 5; in rats: 9,127 ± 2,759 µm², n = 4; Mann-Whitney U test: P = 0.02, see Fig. S14). Consequently, a larger impedance load (larger “sink”) is expected at the soma in human L2/3 neurons.

To examine the impact of impedance load, we computationally substituted the basal tree of human neurons with the basal tree of rat and vice versa (creating “hybrid cells”). This substitution diminished the inter-species differences in latency and velocity (both in the original units and after normalizing the distance and time in units of λ and τ, respectively). Examples of these “hybrid cells” are shown in Fig. 6G,H. In these cases, the basal trees of the 5 modeled human neurons (blue dots) and the basal tree of “Rat1”, “Rat2” and “Rat3” (red dots) were all replaced with the basal tree of “Rat4” neuron. This resulted in a significant reduction in EPSP velocity in the human neurons, diminishing the differences in signal latency/velocity between humans and rats (average latency in humans: 3.42 ± 0.5 ms or 0.23 ± 0.03 in units of τ, n = 5; in rats: 3.2 ± 0.2 ms, 0.21 ± 0.01 in units of τ, n = 4; Mann-Whitney U test: P = 0.7 for both ms and τ units. Average velocity in humans: 0.06 ± 0.005 m/s or 1.86 ± 0.1 λ/τ, n = 5; in rats: 0.05 ± 0.005 m/s or 1.8 ± 0.1 λ/τ, n = 4; Mann-Whitney U test: P = 0.73 for m/s units and P = 0.66 for λ/τ units. See Fig.S13 and Suppl. Table 3). Repeating this procedure for all modeled PCs, but now the basal of any given cell (of both human and rat) was replaced, one-by-one, by the basal tree of all other cells. Again, this confirmed that the significant difference in EPSPs latency between humans and rats consistently diminished and became insignificant in the “hybrid cell” manipulations. More specifically, human basal trees on rat cells accelerate the EPSPs, whereas rat basal trees on human neurons decelerate the EPSPs (Fig S7).

To further demonstrate the effect of switching the basal tree between human and rat neurons (“hybrid cells”) on the EPSPs’ velocity and latency, we depict in Figure 6I the case where the basal tree of “Human1” PC was replaced with the basal tree of “Rat4” (top left) and vice versa (lower right). The result (the “latency-gram” of the EPSPs) is depicted in color-code, showing deceleration in the apical tree of the human cell (top left) and acceleration in the rat’s apical tree (lower right) due to these manipulations. Exemplar somatic EPSPs originated at ∼282 µm from the soma in the original cells (black line) and in the respective “hybrid case” (dashed lines) are shown at top right. The deceleration in the case of a human cell with a rat basal tree is shown on the left and the acceleration in the case of a rat cell with a human basal tree is shown on the right. The explanation for the surprising large impact of the impedance load of the basal tree on signal propagation in the apical tree is elaborated in the Discussion.

In addition to morphological features influencing EPSP latency and velocity, the three key passive parameters - specific membrane resistivity (R_m), capacitance (C_m) and axial resistivity (R_a) - affect signal propagation properties in dendrites by altering the cables’ space constant (λ) and membrane time constant (τ). These changes can either enhance or compensate for the increased velocity of signal propagation of human cells, depending on the specific values of these parameters. Thus, we next examined to what extent the cable parameters of the individual PCs studied here influence signal propagation in their respective dendrites. To address this, we fitted the cable parameters for each of the 9 reconstructed PCs individually, based on double-electrodes recordings (soma and dendrite) for each cell. Figure 7A shows an exemplar reconstructed human L2/3 PC (Human5) with the locations of the two recording/stimulating electrodes used for this cell. Figure 7B top (D-S: dendrite-to-soma direction) shows the case where the step current was injected at the dendrite (cyan). The resultant voltage response is depicted in cyan in the trace below; the model fit is superimposed in dark blue. The opposite (S-to-D) direction is depicted by the next three traces below. This fit enabled a direct estimate of the cable parameters per cell (Table 1). We found that R_m is larger in humans and C_m is smaller in humans (Table 1). Yet the membrane time constant (τ = R_m*C_m) is statistically similar in the two species (Table 2 and see Fig. S15).

Figure 7C-F extends the simulations using the fitted (rather than uniform) cable parameters for each cell (Fig. 6). Compared to the uniform case, the differences in EPSP latency and propagation velocity between and within the two species are slightly enhanced (compare Fig. 7C,D to Fig. 6A,B). For the per-cell fit, the latency ranges between 0.1 - 11 ms for rats (red) and 0.1 – 28 ms for humans (Fig 7C); the velocity ranges between 0.02 - 0.085 m/s for rats (red) and 0.02 – 0.75 m/s for humans (Fig 7D). After normalizing the distance by the space and time constants calculated per cell, the differences in both latency (Fig. 7E) and velocity (Fig. 7F) among individual cells is larger compared with the uniform case (Figure 6D,E). Importantly, despite this increased variance within species, the differences between humans and rats remain statistically significant, both with and without normalization (average latency in humans: 2.3 ± 0.4 ms, or 0.19 ± 0.03 τ, n = 5; in rats: 3.2 ± 0.6 ms or 0.27 ± 0.05 τ, n = 4; Mann-Whitney U test: P = 0.03 for both ms and τ units. Average velocity in humans: 0.085 ± 0.009 m/s or 2.5 ± 0.4 λ/τ, n = 5; in rats: 0.05 ± 0.0085 m/s or 1.7 ± 0.3 λ/τ, n = 4; Mann-Whitney U test: P = 0.03 for m/s and P = 0.02 for λ/τ units. See Fig.S13 and Table 2).

Next, we applied our “hybrid cells” method (as in Fig. 6 G-I). As a result, the inter-species differences were diminished (average latency in humans: 2.3 ± 0.4 ms, 0.19 ± 0.03 τ, n = 5; in rats: 3.2 ± 0.6 ms, 0.27 ± 0.05 τ, n = 4; Mann-Whitney U test: P = 0.9 for ms units and P = 1.0 τ units. Average velocity for human: 0.08 ± 0.02 m/s, 2.39 ± 0.2 λ/τ, n = 5; rat: 0.05 ± 0.005 m/s, 1.7 ± 0.4 λ/τ n = 4; Mann-Whitney U test: P = 0.2 for m/s and P = 0.8 for λ/τ units. See Fig. S8, Fig.S13 and Suppl. Table 4).

To rank the impact of the various factors affecting EPSP propagation latency in human and rat neurons, we conducted a comprehensive statistical analysis using two complementary approaches: the generalized linear model (GLM) (Kiebel & Holmes, 2007) as well as SHAP (SHapley Additive exPlanations) (Lundberg & Lee, 2017) based on fitting Gradient Tree Boosting model (Friedman, 2002). We began by fitting a GLM without interaction terms among the factors affecting EPSP latency (Suppl. Table 5). This enables us to quantify the primary individual factors affecting EPSP propagation. Our analysis revealed the following ranking order: 1) physical distance of synapses from soma had the strongest effect; 2) species differences; 3) conductance load, as demonstrated by our “hybrid cells” manipulation; 4) radii of the apical dendrite, affecting the cables’ space constant, λ; and 5) the specific cable parameters, as revealed when using per-cell fitted parameters versus uniform cable parameters, was minimal. We next performed GLM analysis with interaction terms showing that, as expected, there are significant interactions between the factors affecting EPSP latency (Suppl. Table 6). To further validate the above ranking while incorporating the interactions between the various factors affecting EPSP latency, we performed a SHAP analysis. Notably, even with interactions included, the ranking of the factors affecting signal propagation are aligned with the results from the analysis based on the GLM without interaction terms (see Fig S.16).

We summarize this section by noting that our theoretical efforts enabled the dissection of morphological and electrical parameters that affect differences in EPSPs velocity and latency in human versus rat L2/3 PCs’ dendrites. We first assumed uniform cable properties for all cells modeled (Fig. 6), showing that the key parameter affecting the enhanced velocity in human neurons is the large increase in conductance load (sink) imposed by the extended basal tree in human PCs. The larger diameter of the apical dendrite in human also contributes, but to a lesser degree, to this effect. Finally, differences in passive cable properties also slightly favor faster signal propagation in humans. Indeed, when the basal tree of human PCs modeled (now with cable parameters fitted per cell) was replaced by the basal tree of rat PCs (and vice versa), the interspecies differences diminished – emphasizing again the key impact on signal propagation velocity of the large conductance load at the soma of human L2/3 PCs resulting from the larger basal tree in human PCs (Figs. 7 and 8).

Somatic whole-cell recordings were obtained at ∼37°C from simultaneously recorded PC-PC cell pairs visualized by infrared differential interference contrast (DIC) video microscopy at depths 60– 160 µm from the surface of the slice (Zeiss Axio Examiner LSM7; Carl Zeiss AG, Oberkochen, Germany), 40× water-immersion objective (1.0 NA; Carl Zeiss AG, Oberkochen, Germany) equipped with Luigs and Neumann Junior micromanipulator system (Luigs and Neumann, Ratingen, Germany) and HEKA EPC 10 patch clamp amplifier (HEKA Elektronik GmbH, Lambrecht, Germany). Signals were digitalized at 15 kHz and analyzed with custom written scripts in Python. Presynaptic cells were stimulated with a brief suprathreshold current paired pulse (800 pA, 2–3 ms, 50-60 ms separation of the two pulses), derived in 10s interval. The postsynaptic cells were recorded in current-clamp recording, holding current was set to keep the cell’s membrane potential around −50 mV. The experiments were stopped if the series resistance (Rs) exceeded 25 MΩ or changed more than 20%. For the dendritic recordings 20 μM Alexa 594 was added to the internal solution of the somatic pipette and 20 μM Alexa 488 to the internal solution of the dendritic pipette. The PCs were kept in whole cell configuration at least 10 minutes before the axon bleb or dendritic targeted recording started. Then the microscope switched to 2p mode. The fluorescent dyes of the pipette solution were excited at 850 nm wavelength with a femtosecond pulsing Ti:sapphire laser (Mai Tai DeepSee, Spectra-Physics, Santa Clara, CA). The axon blebs and the dendrites were targeted in 2p mode. After the successful seal formation, the imaging was switched off to reduce the phototoxicity in the sample. All the recordings were carried out in current clamp mode. 800ms long square pulses with elevating amplitude (from -110 to 300 pA) were used to evoke APs. In some experiments the same long square injection protocol was repeated at the dendritic/axonal recording site. For measuring the forward propagation of electrical signals in dendrites, we applied either short artificial EPSC-shaped currents (Connelly et al., 2016) or mostly ramp currents (Markram & Sakmann, 1994) through the dendritic pipette. 10 minutes of recording we applied different drugs or finished the recordings. At the end of the recording, we acquired a 2p Z series from the recorded dendrite. Then the pipettes were carefully withdrawn from the cells. The slices went under chemical fixation for further anatomical investigation. Due to the small diameter of the dendrites of L2/3 neurons, the dendritic pipette access resistance was 92.43 ± 34.29 MΩ with 24.8-196.2 MΩ range (Gidon et al., 2020). We ran a set of computer simulations on our reconstructed neurons (both of human and rat), adding a simulated electrode with variable serial resistance values. We found that, for series resistances ranging from 40-200 MΩ, the effect of the presence of the electrode on the EPSP latencies is negligible (Fig S12.)

The specific membrane capacitance recordings were carried out as described previously (Gentet et al., 2000). Briefly, the L2/3 PCs were whole cell patch clamped, and a gentle suction made during slow withdrawal of the pipette. The nucleus of the cells were pulled out and the voltage clamped at -70 mV. -5mV voltage steps (repeated 100 times) were applied and the capacitive transients were measured. A DIC image of the nucleus were made for calculation of the membrane surface with the following equation: Where a is the shorter diameter of the ellipse and b is the longer one. After the recording the nucleus was blown away and the pipette tip was pushed into a sylgard ball until the GΩ seal formed. The - 5 mV voltage steps were applied again to record the residual capacitance of the system. Before the analysis we subtracted the residual capacitance from the transients.

Pharmacological experiments were carried out on PCs during simultaneous somatic and dendritic recordings after 10 minutes of control recording using ACSF with the following drugs: 20 µM 4- (N-ethyl-N-phenylamino)-1,2 dimethyl-6-(methylamino)pyrimidinium chloride (ZD7288) (Sigma-Aldrich), or 1 µM TTX, 200 µM CdCl₂, and 20 µM AP5.

After electrophysiological recordings, slices were fixed in a fixative containing 4% paraformaldehyde, 15% picric acid, and 1.25% glutaraldehyde in 0.1 M phosphate buffer (PB; pH = 7.4) at 4°C for at least 12 hr. After several washes in 0.1 M PB, slices were cryoprotected in 10% then 20% sucrose solution in 0.1 M PB. Slices were frozen in liquid nitrogen then thawed in PB, embedded in 10% gelatin, and further sectioned into slices of 60 µm in thickness. Sections were incubated in a solution of conjugated avidin-biotin horseradish peroxidase (ABC; 1:100; Vector Labs) in Tris-buffered saline (TBS, pH = 7.4) at 4°C overnight. The enzyme reaction was revealed by 3’3-diaminobenzidine tetrahydrochloride (0.05%) as chromogen and 0.01% H₂O₂ as an oxidant. Sections were post-fixed with 1% OsO₄ in 0.1 M PB. After several washes in distilled water, sections were stained in 1% uranyl acetate, dehydrated in an ascending series of ethanol. Sections were infiltrated with epoxy resin (Durcupan, Sigma-Aldrich) overnight and embedded on glass slices. 3D light microscopic reconstructions were carried out using the Neurolucida system with a 100× objective. The number of putative synaptic contacts were determined by searching for close appositions of presynaptic axon terminals and postsynaptic dendrites under light microscopy. The distance of the contact sites alongside the branches were measured with Neurolucida. The intersomatic distance was calculated from the branch length from the presynaptic soma to the putative synaptic contact alongside the axon, and the length of the dendrite from the contact site to the postsynaptic soma. If there were more than one putative synapse between the cells, we took the shortest intersomatic path distance for that given cell pair.

Sample preparations for the electron microscopy were performed as described previously ^2,6. Briefly, digital images of serial EM sections (20 nm thickness) were taken at 64000x magnification with a FEI/Philips CM10 electron microscope equipped with a MegaView G2 camera. The membrane thickness measurements were carried out on digital images with a custom software. Briefly, postsynaptic dendritic structures were identified with the presence of postsynaptic densities (PSD) faced in front of axon terminals filled with vesicles. At least 20 nm away from the PSD, perpendicular lines were used as region interests (ROI). The intensity line profile of each ROI was calculated, and edge detection was carried out on them. The thickness was determined as the distance between the first and last point along the line profile where the gradient magnitude was larger than 50.

The electrophysiological recordings were analysed by custom written python scripts. First the recorded sweeps were exported with HEKA FitMaster to ascii files. The mean synaptic delay in the paired recordings was determined by the averages of the delays between the peak of single presynaptic action potentials and the onsets of the corresponding EPSPs. The onset was determined by the projection of the intersection of two linear fits on the postsynaptic signal (Fedchyshyn & Wang, 2007). The first line was fitted to the baseline 1 ms window from -0.5 to +0.5 ms of the presynaptic AP peak. The second line was fitted on the rising phase of the EPSP (5-30% of the amplitude). The time point of the crossing lines was projected back to the signal and it was used as the onset (Fig. 1B). For the forward propagation dendritic experiments the latency was calculated on an average signal. The onset of the EPSP-like waveform was determined as the onset of EPSPs in the paired recordings.

The bAP latency was measured at the peak of the average signal for each cell (Stuart & Sakmann, 1994). Only the first APs of the sweeps were averaged to avoid activity dependent Na⁺ channel inactivation that can cause a putative modulatory effect on the signal propagation speed. For the axon bleb recordings we assumed that the axon initial segment (AIS) of the cells are 35 µm from the axon hillock (Palmer & Stuart, 2006), and the APs propagate forward (to the bleb) and backward (to the soma) at the same speed. For the correction of the AIS we used the following formula: where vcorr is the corrected propagation speed for AIS position, l is the axonal distance between the soma and the axon bleb, t is the latency between the two measuring point, ais is the assumed position of the AIS alongside the axon (35 µm).

We constructed detailed passive compartmental and cable models for five L2/3 human neurons and the four rat L2/3 neurons that were both 3D morphologically reconstructed and biophysically characterized. For each modeled neuron, we optimized the values of three key passive parameters: the specific membrane resistivity and capacitance (R_m, C_m) and the specific axial resistivity, R_a, using Neuron 8.0 (Hines et al., 2009) principal axis optimization algorithm (Brent, 1976; Segev et al., 1989). Optimization was achieved by minimizing the difference between experimental voltage response following hyperpolarizing current steps either to the soma or to the apical dendrite (Fig 7A,B) and the model response. When possible, experimental data was averaged over repetitions of the same stimulus.

To account for the surface area of spines, we used the spine correction factor (F) of 1.9 and 1.5 for human and rat PCs, respectively, by multiplying C_m and dividing R_m by F in segments which are at a distance of at least 60 μm from the soma (Eyal et al., 2016; Hunt et al., 2022). In this study we did not attempt to fit the nonlinear effect of I_h of the voltage response of the cells.

As our experimental data contains simultaneous soma-dendritic pair recordings/stimulation, we decided to fit the voltage response in one location (e.g., the soma) for the current injection in the other location (e.g., dendrites). This is a cleaner way compared to the typical case when only one recording/stimulating electrode is available, as the problem of bridge balance at the current input site does not exist in this case. As we have two recording and simulation sites, we also examined how well the model predicts the local voltage response at the injection site (Fig 7B). Analysis and simulation were conducted using Python 3.8 and visualization using matplotlib 3.15 (Hunter, 2007) and seaborn 0.11 (M. Waskom, 2021).

We used the NEURON simulator (Hines et al., 2009) to model the nine 3D reconstructed neurons (Fig. S6). To compute EPSP’s propagation latency and velocity, we simulated EPSPs by injecting a brief transient alpha-shaped current, 𝐼_𝛼 , delivered either to the soma or in various dendritic loci along the modeled apical tree. where 𝐴 = 1.5, 𝜏₀ = 0.25 and 𝜏₁ = 1𝑚𝑠, resulting in EPSP peak time, 𝑡_{𝑝𝑒𝑎𝑘} = 0.5𝑚𝑠 and peak current of 𝐼_{𝑝𝑒𝑎𝑘} = 1.4𝑛𝐴.

Latency of the resultant EPSP was calculated as the difference between the EPSP peak at all dendritic branches and its resulting EPSP at the soma; using a sampling time bin of 0.01ms. Velocity was calculated as the distance of the input site from soma divided by latency between these two points. Each dot in Figures 6 and 7 is the respective value for a specific dendritic segment in a specific branch of a neuron’s apical tree. For each measured feature (radius, and velocity), an inset (zoom-in) matching the experimental distance range was added. It shows the average value across dendritic branches with a given distance from the soma, as a function of distance from soma, smoothed with a rolling 10 μm window. For normalizing the path distance of a given dendritic site to the soma in cable units, we calculated the space constant for each dendritic segment (where d is the segment’s diameter). We then summed up the cable lengths of all segments along the path from the dendritic location to the soma. Time was normalized by the membrane time constant τ = R_m*C_m. Note that, for segments far enough from cable boundary conditions and stimulus location, velocity approximately equals the theoretical value of 2λ/τ, (Agmon-Snir & Segev, 1993) see Fig. S5). Hence, in the uniform case where all specific parameters are equal for all cell modeled (Fig 6), normalizing the distance in cable should equalize latency/velocity differences resulting from diameter differences.

To account for brain tissue shrinkage due to fixation, for every segment, diameter and length were scaled based on an estimation of specific neuron shrinkage (see Suppl. Table 1). To account for unequal dye spread, for a few manually picked segments, diameter value was fixed to be equal to its nearby segment (to avoid sudden diameter jump).

“Equivalent cables” for the respective 9 modelled human and rat cells was based on Rall’s cable theory (Rall, 1959). The variable diameter, 𝑑_𝑒𝑞(𝑋), of this cable as seen from the soma is, where X is the cable (electrotonic) distance from the soma and 𝑑_𝑗(𝑋) is the diameter of the j^th dendrite at the distance X from that point of interest. Figure 8A shows such equivalent cables as seen from the soma. The equivalent cable for the basal tree is depicted in red and for the apical tree in blue. This enables one to graphically appreciate the large difference in the conductance load (current sink) imposed by basal tree between human and rat L2/3.

Statistical analyses were performed in Python v.3.6, using the Python packages DABEST (Ho et al., 2019), scipy, numpy, matplotlib (Hunter, 2007), seaborn (Waskom, 2021), pandas, pinguin (Vallat, 2018) and scikit-learn. SHAP (Lundberg & Lee, 2017) and GLM (Kiebel & Holmes, 2007) models were done with shap python package with scikit-learn Gradient Boosting Regressor (Friedman, 2002) and with statsmodels.glm with gamma family. Interaction formula: latency ∼ species + distance + radius + (species × fitted × distance) + (species × hybrid × distance) + (species × radius). No interaction: latency ∼ species + distance + hybrid + radius + fitted.

Data presented as the mean ± s.d. Normality was tested with the Shapiro-Wilk test. For statistical analysis, t-test, Mann-Whitney U -test or Wilcoxon signed-rank test was used. Correlations were tested using Pearson’s correlation, respectively. We used the Gardner-Altman estimation plot throughout this study which is a bootstrap-coupled estimation of effect sizes, plotting the data against a mean (paired mean, as indicated) difference between the left-most condition and one or more conditions on the right (right y axis), and compared this difference against zero using 5,000 bootstrapped resamples. In these estimation graphics, each black dot indicates a mean difference, and the associated black ticks depict error bars representing 95% confidence intervals; the shaded area represents the bootstrapped sampling-error distribution (Ho et al., 2019). Differences were accepted as significant if p < 0.05. The complete results of all the statistical analysis presented on the main and supplementary figures can be found as a supplementary table.